Extracting node comparison insights for the interactive exploration of property graphs

Requirements

This code has been tested with Python 3.12, and Neo4J 5.x, to install the required packages we recommend the use of a python virtual environement.

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Using custom datasets

Sample data are provided in this repository. To use your own data, you need to setup the graph in a local Neo4J database (using Neo4J desktop for example) in order to process indicator extraction. We only tested with Neo4J 5.x on Mac OS and Linux.

Indicators extraction

To extract indicators from property graphs, use the collect_indicators.py script. This script uses command line arguments to specify the Neo4J database and parameters:

• -r: the number of runs
• -dh: the threshold for acceptable variance (high)
• -dl: the threshold for acceptable variance (low)
• -c: the threshold for non-redundancy (Pearson's correlation)
• -n: the threshold for acceptable density
• -u: a list of suffixes for discarding properties
• --pushdown: if unwanted properties, acceptable density (validation) are pushed down indicator collection
• --keep-nulls: to remove nodes with at least one null value
• --create-index: to create all indices on numerical properties
• config: the json file for the Neo4j database configuration (see examples in the config subdirectory)

Solving the Partiton/Clustering problem

To run the clustering and indicator partition heuristics, use the main.py script. This script uses command line arguments to specify datasets and parameters:

• --k: the number of clusters desired
• --steps: To limit the local search steps
• --method: ls : local search, exp : full tree enumeration, sls: 'simple' start local search
• --dataset: Use one of : iris (debug only), airports, movies, directors, actors or custom
• --path: the path to the dataset (custom only, must be a csv file with a header)
• --delimiter : for custom dataset